Rain doesn't cause dental no-shows. We checked 38,000 appointments.
· Jack Jia · 7 min read
- xona
- dental
- no-shows
- weather
- production-data
There is a story every front desk in the country tells when the morning’s no-show count is bad: “it’s the weather.” It rained last night. It’s snowing. It’s icy. Patients don’t want to drive in this.
The story is intuitive, and it is wrong. Or at least: it is unsupported by the data we have, which is a non-trivial amount of data.
Source: ~38,000 appointments across anonymized BC general dental practice appointment datasets, January 2024 through May 2026. Each appointment was joined to historical hourly weather at the practice’s coordinates via the Open-Meteo Archive API. “Lost” = patient no-show or cancelled. ±10% windowing where exact numbers would identify a practice.
What we measured
For every appointment in the dataset, we looked up the actual weather at the appointment hour, at the practice’s geographic location. We classified each appointment hour into one of four buckets:
- Dry — no precipitation in that hour
- Rain — light to moderate rainfall (0.1–2.5mm/h)
- Heavy rain — over 2.5mm/h
- Snow — any measurable snowfall
We also separately classified entire days as storm-days (top-decile rainfall total for the year, or any snow) versus calm days, to capture the case where the rain was earlier in the day or driving conditions were bad on the way to the appointment.
The result
Across the dataset, lost-rate by weather bucket at the appointment hour:
| Weather at the appointment hour | Lost rate |
|---|---|
| Dry | ~11.3% |
| Rain | ~10.6% |
| Heavy rain | ~10.7% |
| Snow | ~8.1% |
And at the day level:
| Day type | Lost rate |
|---|---|
| Calm day | ~11.0% |
| Storm day (top-decile rain or any snow) | ~10.8% |
These differences are inside the noise floor. Rain is, if anything, very slightly better than dry — but not meaningfully. Snow appears better still, but the snow sample is small and skews toward weekday wintertime hours where the population of bookings was already filtered.
We did not find any combination of slicing — slot × weather, weekday × weather, appointment-type × weather, storm-day vs calm — where the weather effect on lost-rate exceeded a percentage point.
The folk theory of weather-driven no-shows does not survive the data.
Why the intuition feels true anyway
The reason front-desk staff are convinced rain causes no-shows is the same reason most folk explanations stick: availability. On rainy mornings when the no-show count happens to be high, the rain is salient and provides a ready explanation. On dry mornings when the count is just as high, no one says “it’s the lack of weather” — they say “it’s just a bad day.”
Memory selectively pulls out the cases that fit the story. The data, which includes both the rainy bad days and the dry bad days and the rainy good days and the dry good days, does not.
The same psychological bias applies in reverse: the dentist who installs a fancy reminder system in January and watches no-shows drop in March attributes it to the reminder. But January is short-staffed flu season; March is post-winter recovery. The seasonal baseline alone might explain the change. Confounded variables and small samples create the illusion of effects that aren’t there.
What this means in practical terms
A few uses of this finding:
Stop staffing up the front desk for rain. If the practice has a policy of bringing in extra confirmation calls or short-notice patches on rainy days, the data says it’s wasted effort. The no-shows on rainy days are coming whether or not it’s raining; they are coming because of patient-history and slot-level factors that have nothing to do with the weather.
Stop blaming the weather in the post-mortem. When the no-show count is high on a Tuesday morning, the cause is almost certainly who got booked into Tuesday morning, not what the sky was doing at the time. The previous posts in this series describe what those who-and-when factors actually are: a small minority of high-loss patients, plus structural penalties on certain slots (notably 8am) and certain appointment types (notably recall).
Do not overfit reminder work to weather. When evaluating schedule-risk signals, an “is-it-raining” rule looks tempting because it matches front-desk intuition. The data above argues against it. The stronger signals are patient history, appointment type, lead time, and slot risk.
Why this is the kind of finding worth running the query for
Folk theories are sticky because they’re cheap. Anyone can generate one, no one is liable for them being wrong, and they fit the available evidence as long as no one queries the rest of the evidence.
A practice that queries its own data periodically — even with very basic tools — discovers which folk theories work and which don’t. The weather theory is one example of one that doesn’t. The next post in this series, on age and patient demographics, is going to retire a couple more.
What you would need to disconfirm this
If you wanted to argue the data is missing something:
- Severe extreme weather (windstorm, ice storm, atmospheric river). The dataset includes regional storm and snow days that fell inside the January 2024 through May 2026 window. The 8.1% snow-day rate above includes those events. The result is unchanged.
- Patient subgroups (elderly, disabled, far-from-clinic). We haven’t yet sliced by patient distance to clinic. It’s possible a far-from-clinic patient is meaningfully more weather-sensitive than the average — a future post will address this with FSA-level distance data. The headline at the population level is unchanged regardless.
- Specific catastrophes that closed roads. When the practice itself closed because of weather, those appointments don’t appear in the dataset (they were rescheduled by the practice, not lost by the patient). So the rate we are reporting is “patient lost it” under conditions where the practice was open and operational.
The honest sentence
Across tens of thousands of appointments and 28 months of joined hourly weather data, the rain-causes-no-shows theory does not appear in the numbers. The cause of no-shows is somewhere else entirely. Stop chasing the weather.
The interesting practical question for any practice is: if not weather, then what? This series has so far answered that with three findings — patient-history concentration, recall fragility, and time-of-day asymmetry. Together those three explain far more of the schedule risk we can see. None of them are the weather.
This is the fourth post in a short series on what 28 months of BC dental data actually says about no-shows. The remaining followups will look at patient distance from the clinic, the interaction between age and recall behaviour, and how the same dataset can be used to predict next-week risk for any individual patient.
If you want to know what actually predicts your practice’s no-shows — patient history, lead time, slot risk — the free Dental Leakage Scan is where we start.